Room: AAPM ePoster Library
Auto-delineating gas regions on daily MRI for MR-guided online adaptive radiotherapy (MRgOART) of abdomen is challenging since the gas regions occur randomly and their MR intensities can be similar to other tissue types. A novel method is presented to auto-delineate gas regions on MRI considering uncertainty of deformable image registration (DIR).
The proposed method consists of (1) populating bowel region (including bowel, stomach, duodenum)from reference to daily MRIs by DIR, (2) modifying the bowel region along high DIR uncertainty, and (3) generating gas regions in this region using pre-determined threshold. The high DIR uncertainty region was calculated by metric G, defined as the absolute difference between the deformed daily and the reference images, divided by the gradient of the reference image. By analyzing deformed contours and testing with synthetic deformation fields, G was found to be a better predictor of DIR errors than other metrics (e.g. absolute difference or Jacobian determinant). This method was tested on 8 abdominal daily MRI sets collected using 5 different MR sequences with different motion characteristics on a 1.5T MR-Linac. The dice coefficient (DC) was used to compare manually-generated gas volumes with the automatically-delineated volumes using the proposed method and with deformed volumes with uniform expansions. Dose volume histograms (DVH) were compared between the plans calculated with gas density (0.01g/cc) override in the manual vs auto-generated gas regions.
The proposed method is fully automated and executed within 1 minute. The DC of manual and auto-generated gas contours was >95% for all cases, while it was 75%, 83%, and 85% for 0, 0.5 and 1 cm uniform expansion. The DVHs for auto gas contours were virtually identical to those for manual gas override.
The proposed method can quickly and automatically delineate gas regions on daily MRI for MRgOART of abdominal tumors.